voltage sense circuit with selectable range and protection

[wergor]

Hi,
for my current project, I need a high accuracy voltage sense circuit with selectable ranges. I have come up with a simple circuit:

SENSE_HIGH and SENSE_LOW are exposed for the user to connect to a DUT. SENSE_HIGH can assume voltages of up to +-30V in normal operation (but the device should endure much higher voltages being present). SENSE_LOW is connected to ground via a current sense shunt resisitor and will stay within +-30mV of GND in normal operation. R801 is a decade resistor divider, dividing the voltage at SENSE_HIGH in steps of 1/10, 1/100, 1/1000 and 1/10000. U802 is an instrumentation amplifier with a fixed gain of 100. A range is selected by closing one of the relays K801 to K805. D806 clamps U802's input to within ~300mV of the rails, R801 limits the current through D806 to a value it can safely handle. The voltage at sense_volt_+5V is fed into a level shifter and from there to an ADC.

I want to achieve an accuracy of at least 100ppm (after calibration). Is this circuit ok? Are there better designs I should look at? I'm a bit worried that D806's reverse leakage current could introduce a (substantial) error. The LTC1100 datasheet has no information about ESD protection diodes so I'm afraid I cannot simply omit D806.

[David Hess]

The highest output resistance of the divider is about 1 megohm.  The input bias current from the instrumentation amplifier could be up to 65 picoamps which would create an error of 65 microvolts or slightly more than 200ppm on the 0.3 volt range so you have already more than doubled your error budget with the wrong amplifier.  (1) Chopper stabilized amplifiers are not particularly low input current and current noise would also limit low frequency precision.

Leakage through the schottky diodes used for input protection could be 10 thousand times worse.

So to address the schottky diodes first, add a series resistance before the diodes, 10s to 100s of kilohms is typical, to limit current, replace the diodes with low leakage diodes, and maybe add a lower value series resistor at the amplifier input for additional protection.  2N4117 JFETs work well as tested low leakage diodes.  Or with the series resistance, the LTC1100 internal diodes may be enough.

If the series resistance is made about 1 megohm, then it will reduce the output resistance change to 2:1 somewhat reducing the effect of the input bias current (2) but at the cost of raising input noise.  Bypass the 1 megohm series resistance with about 1000 picofards to control noise.

(1) The safe thing to do is be pessimistic about errors caused by input current.  Typically in these types of circuits, an even lower input bias current amplifier is used with automatic zero to remove offset voltage errors instead of chopper stabilization.  Chopper stabilized amplifiers are not suited to high impedance circuits for a variety of reasons.  A precision JFET amplifier or sometimes low input bias current bipolar amplifier is usually a better choice.

(2) This assumes that the input bias current is constant over the input voltage range which may not be the case but adding the 1 megohm resistor may be worth trying.

[MarkF]

A ULN2003A instead of all the MOSFETs would simplify your circuit.
And would be compatible with any 5V MCU.

   

[wergor]

@David Hess Thank you for your feedback! I have updated my circuit:

Edit: I just noticed I forgot to add the bypass capacitor
Here, R802 serves two purposes: first, it limits the current that can flow through the 2N4117 (< 1mA even with 300V between SENSE_HIGH and SENSE_LOW), second, it extends the measurement range a bit, providing approx. 5% overrange.
Can you elaborate on the 2:1 output resistance change? To me it seems as if this would reduce the difference in input bias current related errors between ranges, but not actually reduce the error with any range.

Another option I was thinking about is this:

A 1pA Ib max opamp buffers the output voltage of the resistor divider. Unfortunately, low input bias current opamps tend to have high input offset voltages (max. 160uV in this case) which are also dependent on common mode input voltages. I'm afraid this approach does not leave me in a better position than what I started in.

@MarkF thanks for the tip I have used the ULN200x before but somehow forgot that they existed I'll include them in my design.

[David Hess]

Unfortunately, low input bias current opamps tend to have high input offset voltages (max. 160uV in this case) which are also dependent on common mode input voltages. I'm afraid this approach does not leave me in a better position than what I started in.

Usually what matters is offset voltage drift over temperature because the offset voltage can be nulled out once.

High impedance inputs, like in multimeters, usually do *not* use chopper-stabilized operational amplifiers because of excessive input current and current noise.  What they use instead is an automatic-zero amplifier which periodically shorts the input, measures the offset, and then applies that correction to the following measurements.  This happens between measurements instead of during measurements as with a chopper-stabilized amplifier.

Alternatively, a precision low input bias current amplifier can be trimmed using the offset null pins.  With the best precision parts, that can get within spitting distance of chopper performance and other factors like thermocouple errors become important.

There are some lower input current chopper-stabilized operational amplifier though like the LTC1050, LTC1052, and AD8538 which might squeak by.  I would avoid the LTC1047 because of higher noise.

Do not forget that high common mode rejection is required as well.  If this cannot be met, then the supply voltages can be bootstrapped to improve it.  High common mode rejection is important for precision.